sign in sign up
<<
Home , Antananarivo, Republic of the Congo

Letters https://doi.org/10.1038/s41591-021-01234-8

Variation in SARS-CoV-2 outbreaks across sub-Saharan

Benjamin L. Rice 1,2 ✉ , Akshaya Annapragada 3, Rachel E. Baker 1,4, Marjolein Bruijning1, Winfred Dotse-Gborgbortsi 5, Keitly Mensah 6, Ian F. Miller 1, Nkengafac Villyen Motaze7,8, Antso Raherinandrasana9,10, Malavika Rajeev1, Julio Rakotonirina9,10, Tanjona Ramiadantsoa11,12,13, Fidisoa Rasambainarivo1,14, Weiyu Yu 15, Bryan T. Grenfell1,16, Andrew J. Tatem 5 and C. Jessica E. Metcalf 1,16

A surprising feature of the SARS-CoV-2 pandemic to date is https://coronavirus.jhu.edu/data/mortality). However, variation the low burdens reported in sub-Saharan Africa (SSA) coun- in reporting between countries and some seroprevalence surveys tries relative to other global regions. Potential explanations that suggested high rates of local infection5–7 make it unclear if the (for example, warmer environments1, younger populations2–4) relatively few reported cases and deaths to date indicate a generally have yet to be framed within a comprehensive analysis. We reduced epidemic potential in SSA8. synthesized factors hypothesized to drive the pace and burden Comparisons across SSA populations based on reported infec- of this pandemic in SSA during the period from 25 February tion rates are obscured by heterogeneity in surveillance capacity (for to 20 December 2020, encompassing demographic, comor- example, variation in testing rates among countries) and correlation bidity, climatic, healthcare capacity, intervention efforts and between surveillance and infection reporting9 (Extended Data Fig. 1). human mobility dimensions. Large diversity in the probable Combining reported death counts with assumptions about the prob- drivers indicates a need for caution in interpreting analyses ability of mortality given infection2 yields generally low estimates of that aggregate data across low- and middle-income settings. the percentage of the population expected to have been infected (that Our simulation shows that climatic variation between SSA is, <10%) but this varies more than tenfold between SSA countries population centers has little effect on early outbreak trajecto- and, critically, is sensitive to assumptions about the death report- ries; however, heterogeneity in connectivity, although rarely ing rate (Fig. 1a) and infection fatality ratio (IFR; Fig. 1b). Serology considered, is likely an important contributor to variance in provides an alternative and more direct measure of the percentage the pace of viral spread across SSA. Our synthesis points to infected. Initial serological studies of blood banks in (5–10%)5, the potential benefits of context-specific adaptation of sur- health care workers in urban (9–16%)7 or from State veillance systems during the ongoing pandemic. In particular, in (20–30%)6 indicate that infection rates could be higher characterizing patterns of severity over age will be a priority in some settings, but only the latter was designed as a representative in settings with high comorbidity burdens and poor access to sample and serology-based estimates are sparse in SSA. care. Understanding the spatial extent of outbreaks warrants Given limitations in inference from direct measures of infection emphasis in settings where low connectivity could drive pro- and death rates, experience from locations where the pandemic longed, asynchronous outbreaks resulting in extended stress has progressed more rapidly provides a valuable basis of knowl- to health systems. edge to assess the relative risk of populations in SSA and identify The trajectory of the SARS-CoV-2 pandemic in SSA is uncer- those at the greatest risk. For example, individuals in lower socio- tain. To date, reported case counts and mortality in SSA have economic settings have been disproportionately affected in high lagged behind other geographical regions. All SSA countries, with latitude countries10,11, indicating poverty as a determinant of risk the exception of and , reported fewer than of increased severity of disease. Widespread disruptions to routine 100,000 total cases and fewer than 1,800 deaths as of December health services have been reported12–14 and are likely to contribute 2020 (Supplementary Table 1)—totals far lower than observed in to the burden of the pandemic in SSA15. The role of other factors, Asia, and the Americas (Africa Centres for Disease Control from demography2–4 to health system context16 and intervention and Prevention (Africa CDC) COVID-19 daily updates https://afri- timing17,18, is also increasingly well-characterized. A summary of cacdc.org/covid-19/, Johns Hopkins Coronavirus Resource Center the main findings and limitations of the study is shown in Table 1.

1Department of Ecology and Evolutionary Biology, Princeton University, Princeton, NJ, USA. 2Madagascar Health and Environmental Research, , . 3Johns Hopkins University School of Medicine, Baltimore, MD, USA. 4Princeton Environmental Institute, Princeton University, Princeton, NJ, USA. 5WorldPop, School of Geography and Environmental Science, University of Southampton, Southampton, UK. 6Centre Population et Développement (CEPED), Institut de Recherche pour le Développement (IRD) and Université de Paris, Inserm ERL 1244, Paris, . 7Centre for Vaccines and Immunology, National Institute for Comnmunicable Diseases, National Health Laboratory Service, Johannesburg, South Africa. 8Department of Global Health, Faculty of Medicine and Health Sciences, Stellenbosch University, , South Africa. 9Faculty of Medicine, University of , Antananarivo, Madagascar. 10Teaching Hospital of Care and Public Health Analakely, Antananarivo, Madagascar. 11Department of Life Science, University of , Fianarantsoa, Madagascar. 12Department of Mathematics, University of Fianarantsoa, Fianarantsoa, Madagascar. 13Department of Integrative Biology, University of Wisconsin-Madison, Madison, WI, USA. 14Mahaliana Labs SARL, Antananarivo, Madagascar. 15School of Geography and Environmental Science, University of Southampton, Southampton, UK. 16Princeton School of Public and International Affairs, Princeton University, Princeton, NJ, USA. ✉e-mail: [email protected]

Nature Medicine | VOL 27 | March 2021 | 447–453 | www.nature.com/naturemedicine 447 Letters NATURE MEDICInE

a b

100.00 100.00 50.00 50.00

∆ ∆

10.00 ∆ 10.00 ∆

5.00 ∆ 5.00 ∆

1.00 1.00

0.10 0.10 to date given death reporting to date given IFR by age curve

IFR by age shift (years) 0.01 0.01 –10 –5 Expected cumulative percentage of the population infected Death reporting rate (5) Expected cumulative percentage of the population infected –2 100 0 50 +2 25 +5 10 +10 Mali Togo Niger Niger Chad Benin Kenya Kenya Liberia Gabon Ghana Malawi Malawi Angola Nigeria Nigeria Guinea Zambia Djibouti Burundi Gambia Gambia Uganda Lesotho Ethiopia Somalia Ethiopia Namibia Rwanda Senegal Eswatini Mauritius Comoros Tanzania Botswana Zimbabwe Mauritania Cameroon Cape Verde South Africa South Africa Madagascar Madagascar Cote d lvoire Cote d lvoire Mozambique Sierra Leone South Burkina Faso Guinea Guinea Bissau Equatorial Guinea of the São Tomé and Príncipe São Tomé and Príncipe Central African Republic Democratic Republic of the Congo Democratic Republic of the Congo

Fig. 1 | Variation in the cumulative percentage of the population infected in SSA countries as expected from reported mortality totals. a,b, The expected percentage of a country’s population infected given the number of reported deaths to date, country-specific age structure and a range of death reporting completeness scenarios (a), or a range of IFR scenarios (b). The global IFR age curves were fitted to existing age-stratified IFR estimates (Methods and Supplementary Table 4) and shifted toward younger or older ages by the specified number of years to simulate higher or lower IFRs, respectively (b). Conservatively, we assumed no variation in infection rates by age. (Infections skewed toward older age groups would result in a higher average IFR and thus a lower expected percentage of the population infected for a given number of deaths.) Reported case and death counts are current as of December 2020 (sourced from the Africa CDC; Supplementary Table 1). Data from and the are not shown as they have zero reported deaths as of December 2020. Comparisons to serological surveys (unfilled triangles) available from blood banks in Kenya5, health care workers in urban Malawi7 and a subnational cluster-stratified random sample from Niger State in Nigeria6 are shown.

Characterizing and anticipating the trajectory of ongoing out- result of previous infection by widespread circulating coronaviruses breaks in SSA requires considering variability in known drivers and is a possibility. how they might interact to increase or decrease risk across popula- The highly variable social and health contexts of countries in SSA tions in SSA and relative to non-SSA settings (Fig. 2). For example, will drive location-specific variation in the magnitude of the burden, while most countries in SSA have a relatively young population the time course of the outbreak and options for mitigation. In this age structure, suggesting a decreased burden (since SARS-CoV-2 study, we synthesized the range of factors hypothesized to modulate morbidity and mortality increase with age2–4), prevalent infec- the potential outcomes of SARS-CoV-2 outbreaks in SSA settings tious and noncommunicable comorbidities could counterbalance by leveraging existing data sources and integrating new SARS-CoV- this apparent demographic advantage16,19–21. Similarly, SSA coun- 2-relevant mobility and climate transmission models. Data on tries have health systems that vary greatly in their infrastructure direct measures and indirect indicators of risk factors were sourced and dense, resource-limited urban populations could have fewer from publicly available databases including from the Health options for social distancing22. Yet, decentralized, community- Organization (WHO), , Population based health systems that benefit from past experience with Division, Demographic and Health Surveys (DHS-USAID), Global epidemic response (for example, to Ebola23,24) can be mobilized. Burden of Diseases and WorldPop, and newly generated datasets Climate is frequently invoked as a potential mitigating factor (Extended Data Fig. 2 and Supplementary Table 3). We organized for warmer and wetter settings1, including SSA, but climate our assessment around two aspects that will shape national out- varies greatly between population centers in SSA and the reality comes and response priorities in the event of widespread outbreaks: of the existence of large susceptible populations could counteract (1) the burden, or expected severity of the outcome of an infection, any climate forcing during initial phases of the epidemic25. which emerges from age, comorbidities and health systems func- Connectivity, at international and subnational scales, also varies tioning; and (2) the rate of spread within a geographical or pace greatly26,27 and the time interval between viral introductions and the of the pandemic. onset of interventions, such as lockdowns, will modulate the tra- We grouped factors that might drive the relative rates of these jectory9. Finally, burdens of , infectious diseases and two features (mortality burden and pace of the outbreak) along six many other underlying health conditions are higher in SSA than dimensions of risk: (1) demographic and socioeconomic param- in other regions (Supplementary Table 2) and their interactions eters related to transmission and burden; (2) comorbidities relevant with SARS-CoV-2 are, as of yet, poorly understood; conversely to burden; (3) climatic variables that may impact the magnitude cross-protection from either SARS-CoV-2 infection or disease as a and seasonality of transmission; (4) prevention measures deployed

448 Nature Medicine | VOL 27 | March 2021 | 447–453 | www.nature.com/naturemedicine NATURE MEDICInE Letters

Table 1 | Policy summary Category Description Background As the SARS-CoV-2 pandemic expanded globally, reported incidence and mortality remained low in SSA. Yet, a general conclusion that SSA may avoid the high burdens seen elsewhere neglects considerable national and subnational variability in likely drivers of the pandemic’s impacts, from its burden to its pace, and does not address variable surveillance and registration systems. Main findings and Synthesizing data on the likely drivers of the pace and burden of SARS-CoV-2 in SSA reveals extensive variability in factors that limitations can define the burden once individuals have become infected. Pairing this with simulations of the trajectory of the outbreak indicates little effect of climate but potentially prolonged outbreaks in many settings due to heterogeneities in connectivity, an effect that could be amplified by control efforts. However, although we provide a qualitative overview of the continued potential impact of the pandemic in SSA, quantitative projections remain intractable given a lack of information on the quantitative impact of important risk factors (from how comorbidities might shape the IFR to how remoteness will reduce spread). Additionally, uncertainties associated with existing surveillance and mortality registration data impede direct comparison of expectations with national data. Policy implications Strengthening surveillance and registration is necessary to narrow the range of expectations for country trajectories in SSA for incidence and mortality. Additional tools for surveillance, such as serology, or approaches that quantify excess mortality, will provide important complementary measures. Our national and subnational analyses point to where returns on investment in strengthening surveillance could yield the greatest returns. Countries with high comorbidity risk may have most to gain from understanding determinants of mortality; low connectivity countries will benefit from investments in delineating the spatial extent of outbreaks; all countries will benefit from evaluating the intersection between epidemic pace and health system disruptions.

Relative SARS-CoV-2 epidemic risk

Direct burden Indirect burden Epidemic pace (SARS-CoV-2 mortality) (disruptions to routine health services)

Mortality rate Viral Transmissibility (IFR) introductions (Rt) Fewer medical facilities on average in SSA Fewer physicians and medical staff on average in SSA Higher infectious disease incidence in SSA Noncommunicable diseases: higher mortality burden per age group in SSA Noncommunicable diseases: lower overall prevalence in SSA Age: younger populations on average in SSA Lower internal connectivity between subregions in SSA Greater mean travel time to cities in some SSA countries Lower frequency of international air travel in SSA Less access to handwashing and other prevention measures Larger size of households for ages >50 in some SSA settings Dense, resource-limited urban populations in many SSA cities Community-based decentralized health systems Recent experience with out- outbreak response (e.g. ) Higher temperature and humidity on average in SSA

E E B B B A F F F D A A D D C

Risk A Demography B Comorbidity C Climate D Prevention E Access to care F Human mobility and travel dimension:

Fig. 2 | Hypothesized modulators of relative SARS-CoV-2 epidemic risk in SSA. The factors (A–F) hypothesized to increase (red) or decrease (blue) mortality burden or epidemic pace within SSA, relative to global averages, are grouped into six categories or dimensions of risk. In this framework, epidemic pace is determined by person-to-person transmissibility, which can be defined as the time-varying effective reproductive number, Rt, and introduction and geographical spread of the virus via human mobility. SARS-CoV-2 mortality (determined by the IFR) is modulated by demography, comorbidities (for example, noncommunicable diseases) and access to care. Overall burden is a function of direct burden and indirect effects due to, for example, socioeconomic disruptions and disruptions in health services, such as vaccination and infectious disease control. Supplementary Table 2 contains the details and references used as a basis to draw the hypothesized modulating pathways. to reduce transmission; (5) accessibility and coverage of existing basic handwashing (that is, clean water and soap28) among urban healthcare systems to reduce burden; and (6) patterns of human households in Mali, Madagascar, Tanzania and Namibia (62–70%) mobility relevant to transmission (Supplementary Table 2). exceed the global average (58%) but are <10% for Liberia, Lesotho, National scale variability in SSA among these dimensions of Democratic Republic of the Congo and Republic of Guinea-Bissau risk often exceeds ranges observed across the globe (Fig. 3a–d (Fig. 3d). Conversely, the range in the number of physicians is low and Extended Data Fig. 3). For example, estimates of access to in SSA, with all countries other than Mauritius below the global

Nature Medicine | VOL 27 | March 2021 | 447–453 | www.nature.com/naturemedicine 449 Letters NATURE MEDICInE

a SSA AMR EMR EUR SEA WPR e Max

800 MUS 200 LBR 60 600 150 SWZ Age

LSO > 50 years) 400 MLI (percentage of the population SYC CAF 100 ZAF Min Physicians CPV AGO Min 200

50 GNQ NAM NGA BWA GAB NAM (per 100,000 population) STP

40 size Mean

AGO CIV COG DJI KEN GNB MDG COM MRT GHA BEN RWA MLI COG GMB

CMR COD UGA ZMB ETH GAB household ZWE SWZ 0 SEN MOZ GIN Max CAF LSO ERI BFA TCD TGO BDI NER LBR TZA (> 50 years) SOM SLE MWI 0 COD

b SSA AMR EMR EUR SEA WPR SLE

LBR BFA SEN COM ZAF 60 TCD MOZ SWZ BDI CIV LSO MLI GNQ

CAF MWI BEN

100 AGO MDG 20 NER GIN ETH GHA

disease mortality (per 100,000 population) TGO CMR UGA NGA NAM

40 COG GAB

COD TZA Age-standardized chronic obstructive pulmonary RWA SLE CPV

ZAF ZMB GNB ZWE BFA SEN CIV TCD COM

50 BEN MOZ ETH MWI GIN BDI MDG MER ERI GMB SSD CMR UGA TGO DJI NGA SOM

GHA KEN 20 MRT TZA

BWA 0 RWA ZMB KEN Age-standardized chronic 0 25 50 75 100 MUS obstructive pulmonary disease

mortality (per 100,000 population) 0 0 Physicians (per 100,000 population)

SSA AMR EMR EUR SEA WPR c 100

100 SSD CAF TCD STP f MDG GNB MOZ MRT MDG

SLE MiLI COD 75 SOM ETH

75 NER CPV COM ERI GNQ MWI LBR BFA DJI BEN BWA BDI CIV KEN MLI

AGO TZA ZMB UGA

RWA NAM

TGO 60 TZA LSO NGA

50 50 COG GIN SEN CMR

GHA ZAF

GAB COG GMB NAM SWZ SWZ NGA ZWE 25 25 ZAF GHA low-quality housing (%) SEN 40 Urban population living in crowded, 0 0 UGA 80 AGO SSA AMR EMR EUR SEA WPR NER d KEN MLI 100 MDG CIV

TZA SLE NAM GIN ZMB

60 COG MRT ETiH

NGA BFA ZAF CAF 75 20 MOZ

ZWE BDI BEN SWZ TGO CMR COM GHA RWA TCD SEN

40 STP MWI UGA 50 AGO

KEN COD NER CIV SLE GNQ GIN ZMB ETH BFA CAF MOZ BDI TGO

Urban population with access to handwashing facilities (%) LSO COM TCD 25 20 BEN MWI CMR RWA GMB SOM 0 LBR GNB COD handwashing facilities (%) LSO LBR Urban population with access to 0 0 25 50 75 100 Urban population living in crowded, low-quality housing (%)

Fig. 3 | Variation among SSA countries in select determinants of SARS-CoV-2 risk. a–d, Right, SSA countries were ranked from least to greatest for each indicator; the bar color shows the population age structure (percentage of the population above the age of 50). The solid horizontal lines show the global mean value and the dotted lines show the mean among SSA countries. Left, The boxplots show the median, the inner bounds correspond to the interquartile range (IQR, 25th to 75th percentiles) and the outer bounds correspond to the 1.5 × IQR, grouped by WHO-defined geographic regions. SSA, Sub-Saharan Africa; AMR, Americas Region; EMR, Eastern Mediterranean Region; EUR, Europe Region; SEA, Southeast Asia Region; WPR, Western Pacific Region (n = 206, 172, 106 and 92 countries with available data for a–d, respectively). e,f, Bivariate comparisons of the variables shown in a,b and c,d, respectively. The dot size shows the mean household size for households with individuals aged over 50, the dashed lines show the median value among SSA countries and the quadrants with the greatest risk are outlined in red (for example, fewer physicians and greater age-standardized chronic obstructive pulmonary disease mortality). See Supplementary Table 3 and Extended Data Fig. 3 for a full description and link to visualization of all variables. average (168.78 per 100,000 population) (Fig. 3a). Yet, estimates are SSA settings. Conversely, an analysis of the reported age-specific still heterogeneous within SSA, with, for example, Gabon estimated death data available from Kenya and South Africa suggested low to have more than 4 times the physicians of neighboring Cameroon IFRs in comparison to non-SSA countries30. Comparison of empiri- (36.11 and 8.98 per 100,000 population, respectively). This disparity cal age profiles of mortality more broadly across SSA is currently is likely to interact with social contact rates among the aged in deter- limited by the small number of total deaths reported to date for mining exposure and clinical outcomes (for variation in household many countries (for example, 33 of 48 SSA countries have reported size, see Fig. 3e,f). Relative ranking across variables is also uneven fewer than 200 total deaths as of December 2020) and incomplete among countries (Extended Data Fig. 4) with the result that this associated age data. Consequently, we used the global IFR by age diversity cannot be easily reduced (for example, the first two princi- estimates and explored the potential effect on mortality of devia- pal components explain only 32.6 and 13.1% of the total variance as tions from the expected baseline IFR in diverse SSA settings. shown in Extended Data Fig. 5), indicating that approaches reliant Small shifts (for example, of 2–10 years of age) in the IFR pro- on a small subset of variables will fail to capture the observed varia- file resulted in large effects on expected mortality for a given level of tion among SSA countries. infection. For example, Chad, Burkina Faso and the Central African To first evaluate variation in the burden emerging from the Republic, while among the youngest SSA countries, have a high prev- severity of infection outcome, we considered how demography, alence of diabetes and low density of hospital beds. Given the age comorbidity and access to care might modulate the age profile of structure of these countries, a slight shift in the IFR by age profile SARS-CoV-2 morbidity and mortality2–4. Subnational variation toward higher mortality in middle-aged groups (for example, ages in the distribution of high-risk age groups indicates considerable 50–60 years) would result in mortality increasing to a rate that would variability, with higher burden expected in urban settings in SSA exceed a majority of the other, relatively older SSA countries at the (Fig. 4a), where density and thus transmission are likely higher29. unshifted baseline (Fig. 4c and Methods). Generally, minor shifts in Comorbidities and access to clinical care also vary across SSA the IFR lead to differences larger than the magnitude of the differ- (for example, for diabetes prevalence and hospital bed capacity, ence expected from differing age structures for countries in SSA. see Fig. 4b). In comparison to settings where previous SARS- Although there is greater access to care in older populations by CoV-2 IFR estimates have been reported, mortality due to some metrics (Fig. 3a; correlation between age and the number of noncommunicable diseases in SSA increases more rapidly with age physicians per capita, r = 0.896, P < 0.001), access to clinical care is (Extended Data Fig. 6) suggesting risk for an elevated IFR in some highly variable overall (Fig. 4d) and maps poorly to indicators of

450 Nature Medicine | VOL 27 | March 2021 | 447–453 | www.nature.com/naturemedicine NATURE MEDICInE Letters

Range in mortality under simulated IFR scenarios a Baseline mortality risk from demographic structure c d e

Indicators of access Indicators of comorbidity IFR +10IFR years +5IFR years +2Baseline yearsIFR IFR –2 yearsIFR –5 years IFR –10 years to care at national level burden at national level Mauritius Mauritius Seychelles Seychelles Cape Verde Cape Verde South Africa South Africa Lesotho Lesotho Djibouti Djibouti Botswana Botswana Eritrea Eritrea

Eswatini Eswatini National-level risk (quartiles): Namibia Namibia Gabon Gabon Ethiopia Ethiopia 10,000 South Sudan South Sudan 1,000 Liberia Liberia 100 Benin Benin Ghana Ghana 10 Mauritania Mauritania 1 São Tomé and Príncipe São Tomé and Príncipe 0.1 Comoros Comoros

Expected mortality <0.1 Madagascar Madagascar (per 30 × km grid) Rwanda Rwanda Senegal Senegal Zimbabwe Zimbabwe Democratic Republic of the Congo Democratic Republic of the Congo Comorbidity versus access to care Sierra Leone Sierra Leone b Togo Togo Republic of the Congo Republic of the Congo Côte d'Ivoire Côte d'Ivoire Guinea Guinea Mozambique Mozambique Somalia Somalia Missing data for an indicator Central African Republic Central African Republic Guinea Bissau Guinea Bissau Cameroon 0.20 Cameroon Nigeria Nigeria Tanzania Tanzania Malawi 0.15 Malawi Kenya Kenya Gambia Gambia 0.10 Equatorial Guinea IFR Equatorial Guinea Niger Niger Chad Chad 0.05 0 Burkina Faso Burkina Faso 0.2 Mali Mali 0.4 Burundi Burundi 0 0.6 Angola Angola 0.8 0 25 50 75 100

Hospital beds Zambia Zambia Age (years) 1.0 Uganda Uganda 0 0.2 0.4 0.6 0.8 1.0 Diabetes 0 50 100 150 200 250 prevalence(%) Hospitals Expected mortality Physicians (per 100,000 population) Hospital beds exposure urban 2.5 Nurses and midwives Diabetes prevalence PM ImmunizationBirths coverage with skilled Tuberculosisstaff incidenceOverweight prevalence

Health expenditure per capita Raised cholesterol prevalence Raised blood pressure prevalence

Chronic obstructive pulmonary disease mortality

Fig. 4 | Variation in expected burden for SARS-CoV-2 outbreaks in SSA. a, Expected mortality in a scenario where cumulative infection reaches 20% across age groups and the IFR curve is fitted to existing age-stratified IFR estimates (Methods and Supplementary Table 4). b, National-level variation in comorbidity and access to care variables, for example, diabetes prevalence among adults and the number of hospital beds per 100,000 population for SSA countries. c, Range in mortality per 100,000 population expected in scenarios where the cumulative infection rate is 20% and IFR per age is the baseline (black) or shifted ±2, 5 or 10 years (gray). Inset, The IFR by age curves for each scenario are shown. d,e, Selected national-level indicators; estimates of reduced access to care (for example, fewer hospitals) (d) or increased comorbidity burden (for example, higher prevalence of raised blood pressure) (e) shown with darker red for higher-risk quartiles (see Extended Data Fig. 4 for all indicators). Countries missing data for an indicator are shown in gray. For comparison between countries, estimates are age-standardized where applicable (Supplementary Table 3). High-resolution maps for each variable and scenario are available at the SSA-SARS-CoV-2-tool for estimating the burden of SARS-CoV-2 in SSA (https://labmetcalf.shinyapps.io/covid 19-burden-africa/). comorbidity (Fig. 4e). Empirical data are urgently needed to assess as handwashing and other non-pharmaceutical interventions (Fig. the extent to which the IFR-age-comorbidity associations observed 3d), population contact patterns including mobility and urban elsewhere are applicable to SSA settings with reduced access to crowding29 (Fig. 3c) and potentially the effect of climatic variation1. advanced care. Yet both surveillance and mortality registration31 are Where countries fall across this spectrum of pace will shape inter- frequently under-resourced in SSA, complicating both evaluating actions with lockdowns and determine the length and severity of and anticipating the burden of the pandemic and underscoring the disruptions to routine health system functioning. urgency of strengthening existing systems24. Subnational connectivity varies greatly across SSA, both between The frequency of viral introduction to each country, likely subregions of a country and between cities and their rural periphery governed by international air travel in SSA32, determines both the (for example, as indicated by travel time to the nearest city with a timing of the first infections and the number of initial infection population over 50,000; Fig. 5c). As expected, in stochastic simula- clusters that seed subsequent outbreaks. The relative importation tions using estimates of viral transmission parameters and mobility risk among SSA cities and countries was assessed by compiling data (Extended Data Fig. 7), a smaller cumulative proportion of the popu- from 108,894 flights arriving at 113 international airports in SSA lation is infected at a given time in countries with larger populations from January to April 2020 (Fig. 5a), stratified by the SARS-CoV-2 in less connected subregions (Fig. 5d and Extended Data Fig. 8); status at the departure location on the day of travel (Fig. 5b). A including non-pharmaceutical interventions reduces this propor- small subset of SSA countries received a disproportionately large tion still further. At the national level, susceptibility declines more percentage (for example, South Africa, Ethiopia, Kenya and Nigeria slowly and more unevenly in such settings (for example, Ethiopia, together contributed 47.9%) of the total travel from countries with South Sudan, Tanzania) due to a lower probability of introduc- confirmed SARS-CoV-2 infections, which likely contributed to tions and reintroductions of the virus locally, an effect amplified by variation in the pace of the pandemic across settings and is con- lockdowns (Extended Data Fig. 9). It is unclear whether the more sistent with those 4 countries together contributing 74.3% of all prolonged, asynchronous epidemics expected in these countries or reported cases in SSA as of 20 December 2020 (refs. 32,33). the overlapping, concurrent epidemics expected in countries with Once local chains of infection are established, the rate of spread higher connectivity (for example, Malawi, Kenya, Burundi) will within countries will be shaped by efforts to reduce spread, such be a greater stress to health systems. Outbreak control efforts are

Nature Medicine | VOL 27 | March 2021 | 447–453 | www.nature.com/naturemedicine 451 Letters NATURE MEDICInE

a International travel by country c Connectivity (population-weighted mean travel time to nearest city) e Seasonal variation in humidity

International travelers from January to April 2020 Population-weighted 2,000,000 travel time (h) <2 200,000 2–4 4–6 6–8 20,000 8–10 0 5 10 15 10–12 >12 Range q (g kg–1) No data/non-SSA area

b International travelers in 2020 by departure location d Connectivity versus proportion infected at 1 year f Infection time series assuming model with climate forcing Number of reported cases 800,000 40 at source location Control: 0% Control: 10% Control: 20% at time of travel 1,000+ 101–1,000 1–100 600,000 20 0 ) 0.10

0 400,000 Antananarivo 0.05 –20 Lomé 200,000 Proportion infected ( I/ N I/N at t = 1 year (% difference from mean)

–40 Ethiopia South Sudan Tanzania Rwanda Eritrea Niger Chad South Africa Sierra Leone Equatorial Guinea Cameroon Democratic Republic of the Congo Mali Mozambique Madagascar Somalia Benin Botswana Senegal Uganda Central African Republic Namibia Guinea Zambia Burkina Faso Angola Republic of the Congo Mauritania Côte d'Ivoire Togo Zimbabwe Ghana Guinea Bissau Liberia Nigeria Eswatini Gambia Kenya Gabon Malawi Burundi São Tomé and Príncipe Djibouti

0 0 Population-weighted mean travel Apr Apr Apr Apr Jan Jan Jan Jan Mar Mar Mar Mar Feb Feb Feb Feb time to nearest city (h) 0 10 20 30 40 50 0 10 20 30 40 50 0 10 20 30 40 50 1 2 3 4 5 South Africa Ethiopia Kenya Nigeria Time (weeks since case count >10 cases)

Fig. 5 | Variation in connectivity and climate in SSA and expected effects on SARS-CoV-2. a, International travelers to SSA from January to April 2020, as inferred from the number of passenger seats on arriving aircrafts. b, For the 4 countries with the most arrivals, the proportion of arrivals by month coming from countries with 0, 1–100, 101–1,000 and 1000+ reported SARS-CoV-2 infections at the time of travel (see Supplementary Table 5 for all others) is shown. c, Connectivity within SSA countries as inferred from average population-weighted mean travel time to the nearest urban area with a population >50,000. d, Mean travel time at the national level and variation in the fraction of the population expected to be infected (I/N) in the first year from stochastic simulations (Methods). e, Climate variation across SSA as shown by seasonal range in specific humidity, q (g kg−1) (max average q − min average q). f, The effect of local seasonality and control efforts (R0 decreases by 0%, that is, unmitigated, 10 or 20%) on the timing of epidemic peaks (max I/N) in SSA cities (with three exemplar cities highlighted in pink; Methods). likely to be further complicated during prolonged epidemics if they pattern of morbidity and mortality over age. Transparent and timely intersect with seasonal events such as temporal patterns in human communication of these context-specific risk patterns could aid mobility29 or other infections (for example, malaria). community engagement in efforts to reduce transmission, help Despite extreme variation among cities in SSA (Fig. 5e), large epi- motivate population behavioral changes and guide existing net- demic peaks are expected in all cities (Fig. 5f), even from our models works of community case management. incorporating interventions and transmission rates that decline in Because the largest impacts of SARS-CoV-2 outbreaks may be response to warmer, more humid local climates (climate-dependent through indirect effects on routine health provisioning, under- variation in transmission rate for coronaviruses inferred from standing how existing programs may be disrupted differently by endemic circulation in the , but robust to parameter acute versus longer outbreaks is crucial to planning resource alloca- value choice; Methods). After accounting for differences in the date tion. For example, population immunity will decline proportionally of introductions, simulated climate forcing generates a maximum with the length of disruptions to routine vaccination programs34, of only 6–7 weeks’ variation in the time to epidemic peaks, with resulting in more severe consequences in areas with prolonged epi- peaks generally expected earlier in more southerly, colder, drier cit- demic time courses. ies (for example, Windhoek and ) and later in more humid, Others have suggested that this crisis presents an opportunity to coastal cities (for example, Bissau, Lomé and Lagos). Reductions in unify and mobilize across existing health programs (for example, transmission due to control efforts, as expected, prolong the time for human immunodeficiency virus (HIV), tuberculosis, malaria to epidemic peak (Fig. 5f and Extended Data Fig. 10). Apart from and noncommunicable diseases)24. Although this might be a pow- these slight shifts in timing, the large proportion of the population erful strategy in the context of acute, temporally confined crises, that is susceptible overwhelm the effects of climate25; earlier sugges- long-term distraction and diversion of resources35 could be harmful tions that Africa’s generally more tropical environment alone may in settings with extended, asynchronous epidemics. A higher risk provide a protective effect1 are not supported by evidence. of infection among healthcare workers during epidemics36,37 could Our synthesis emphasizes striking country-to-country varia- also amplify this risk. tion in the drivers of the pandemic in SSA (Fig. 3), indicating that As evidenced by failures in locations where the epidemic pro- continued variation in the burden (Fig. 4) and pace (Fig. 5) is to gressed rapidly (for example, United States), effective governance be expected even across low-income settings. Since small pertur- and management before reaching large case counts will likely yield bations in the age profile of mortality could drastically change the the largest rewards. Generalizing across SSA is difficult because national-level burden in SSA (Fig. 4), building expectations for the time course and estimates of the effect of intervention policies the risk for each country requires monitoring for deviations in the have varied greatly (Extended Data Fig. 9); however, Mauritius and

452 Nature Medicine | VOL 27 | March 2021 | 447–453 | www.nature.com/naturemedicine NATURE MEDICInE Letters

Rwanda, for example, have reported extremely low incidence thanks 14. Santoli, J. M. et al. Efects of the COVID-19 pandemic on routine pediatric in part to a well-managed early response. vaccine ordering and administration: United States, 2020. MMWR Morb. Mortal. Wkly. Rep. 69, 591–593, (2020). The burden and time course of SARS-CoV-2 is expected to be 15. Roberton, T. et al. Early estimates of the indirect efects of the COVID-19 highly variable across SSA. Simulations show that variation in inter- pandemic on maternal and child mortality in low-income and middle-income national and subnational connectivity are expected to be impor- countries: a modelling study. Lancet Glob. Health 8, e901–e908 (2020). tant determinants of pace, but variability in reporting regimes 16. Walker, P. G. T. et al. Te impact of COVID-19 and strategies for mitigation makes it difficult to compare observations to date with expecta- and suppression in low- and middle-income countries. Science 369, 413–422 (2020). tions (Extended Data Fig. 7). As the outbreak continues to unfold, 17. Pei, S., Kandula, S. & Shaman, J. Diferential efects of intervention timing on critically evaluating this mapping (Extended Data Fig. 8) can focus COVID-19 spread in the United States. Sci. Adv. 6, eabd6370 (2020). surveillance efforts to areas expected to have prolonged epidemic 18. Lai, S. et al. Assessing the efect of global travel and contact reductions to trajectories and high mortality burdens. The emergence and rapid mitigate the COVID-19 pandemic and resurgence. Preprint at medRxiv spread in southern Africa of lineage B.1.351, with multiple spike https://doi.org/10.1101/2020.06.17.20133843 (2020). 19. Nepomuceno, M. R. et al. Besides population age structure, health and other protein mutations including the N501Y mutation associated with demographic factors can contribute to understanding the COVID-19 burden. increased transmission rate in the UK lineage B.1.1.7, indicates the Proc. Natl Acad. Sci. USA 117, 13881–13883 (2020). importance of genomic surveillance of transmission foci in SSA38. 20. Clark, A. et al. Global, regional, and national estimates of the population at Additional immunological surveys and country-specific analyses increased risk of severe COVID-19 due to underlying health conditions in 2020: a modelling study. Lancet Glob. Health 8, e1003–e1017 (2020). of the age profile of mortality are urgently needed in SSA and will 21. Ghisolf, S. et al. Predicted COVID-19 fatality rates based on age, sex, likely be a powerful lens for understanding the current landscape of comorbidities, and health system capacity. BMJ Glob. Health 5, e003094 (2020). population risk39. When considering hopeful futures with the dis- 22. Winskill P. et al. Report 22: Equity in Response to the COVID-19 Pandemic: an tribution of SARS-CoV-2 vaccines, it is imperative that vaccine dis- Assessment of the Direct and Indirect Impacts on Dsadvantaged and Vulnerable tribution be equitable and in proportion with need. Understanding Populations in Low- and Lower Middle-Income Countries (Imperial College London, 2020); https://www.imperial.ac.uk/media/imperial-college/medicine/ factors that both drive spatial variation in vulnerable popula- mrc-gida/2020-05-12-COVID19-Report-22.pdf tions and temporal variation in pandemic progression could help 23. Kapata, N. et al. Is Africa prepared for tackling the COVID-19 (SARS-CoV-2) approach these goals in SSA. epidemic. Lessons from past outbreaks, ongoing pan-African public health eforts, and implications for the future. Int. J. Infect. Dis. 93, 233–236 (2020). Online content 24. Nachega, J. B. et al. From easing lockdowns to scaling-up community-based COVID-19 screening, testing, and contact tracing in Africa: shared Any methods, additional references, Nature Research report- approaches, innovations, and challenges to minimize morbidity and mortality. ing summaries, source data, extended data, supplementary infor- Clin. Infect. Dis. ciaa695 (2020). mation, acknowledgements, peer review information; details of 25. Baker, R. E., Yang, W., Vecchi, G. A., Metcalf, C. J. E. & Grenfell, B. T. author contributions and competing interests; and statements of Susceptible supply limits the role of climate in the early SARS-CoV-2 data and code availability are available at https://doi.org/10.1038/ pandemic. Science 369, 315–319 (2020). 26. Weiss, D. J. et al. A global map of travel time to cities to assess inequalities in s41591-021-01234-8. accessibility in 2015. Nature 553, 333–336 (2018). 27. Tatem, A. J. WorldPop, open data for spatial demography. Sci. Data 4, Received: 3 September 2020; Accepted: 11 January 2021; 170004 (2017). Published online: 2 February 2021 28. World Health Organization and United Nations Children’s Fund Joint Monitoring Programme for Water Supply, Sanitation and Hygiene. Core References Questions on Water, Sanitation and Hygiene for Household Surveys: 2018 Update. https://washdata.org/sites/default/fles/documents/reports/2019-05/ 1. Mecenas, P., Bastos, R. T. D. M., Vallinoto, A. C. R. & Normando, D. Efects JMP-2018-core-questions-for-household-surveys.pdf (UNICEF and of temperature and humidity on the spread of COVID-19: a systematic WHO, 2018). review. PLoS ONE 15, e0238339 (2020). 29. Korevaar, H. M. et al. Quantifying the impact of US state non-pharmaceutical 2. Salje, H. Estimating the burden of SARS-CoV-2 in France. Science 369, interventions on COVID-19 transmission. Preprint at at medRxiv https://doi. 208–211 (2020). org/10.1101/2020.06.30.20142877 (2020). 3. Rinaldi, G. & Paradisi, M. An empirical estimate of the infection fatality rate 30. O’Driscoll, M. et al. Age-specifc mortality and immunity patterns of of COVID-19 from the frst Italian outbreak. Preprint at medRxiv https://doi. SARS-CoV-2 infection in 45 countries. Nature https://doi.org/10.1038/ org/10.1101/2020.04.18.20070912 (2020). s41586-020-2918-0 (2020). 4. Verity, R. et al. Estimates of the severity of coronavirus disease 2019: a 31. Mikkelsen, L. et al. A global assessment of civil registration and vital statistics model-based analysis. Lancet Infect. Dis. 20, 669–677 (2020). systems: monitoring data quality and progress. Lancet 386, 1395–1406 (2015). 5. Uyoga, S. et al. Seroprevalence of anti-SARS-CoV-2 IgG antibodies in Kenyan 32. Gilbert, M. et al. Preparedness and vulnerability of African countries against blood donors. Science 371, 79–82 (2021). importations of COVID-19: a modelling study. Lancet 395, 871–877 (2020). 6. Majiya, H. et al. Seroprevalence of COVID-19 in Niger State. Preprint at 33. Haider, N. et al. Passengers’ destinations from : low risk of novel medRxiv https://doi.org/10.1101/2020.08.04.20168112 (2020). coronavirus (2019-nCoV) transmission into Africa and South America. 7. Chibwana, M. G. et al. High SARS-CoV-2 seroprevalence in health care Epidemiol. Infect. 148, e41 (2020). workers but relatively low numbers of deaths in urban Malawi. Wellcome 34. Takahashi, S. et al. Reduced vaccination and the risk of measles and other Open Res. 5, 199 (2020). childhood infections post-Ebola. Science 347, 1240–1242 (2015). 8. Mbow, M. et al. COVID-19 in Africa: dampening the storm? Science 369, 35. Cash, R. & Patel, V. Has COVID-19 subverted global health? Lancet 395, 624–626 (2020). 1687–1688 (2020). 9. Skrip, L. A. et al. Seeding COVID-19 across sub-Saharan Africa: an analysis 36. Adams, J. G. & Walls, R. M. Supporting the health care workforce during the of reported importation events across 40 countries. Preprint at medRxiv COVID-19 global epidemic. JAMA 323, 1439–1440 (2020). https://doi.org/10.1101/2020.04.01.20050203 (2020). 37. Kilmarx, P. H. et al. Ebola virus disease in health care workers: Sierra Leone, 10. Deng, X. et al. Case fatality risk of the frst pandemic wave of coronavirus 2014. MMWR Morb. Mortal. Wkly. Rep. 63, 1168–1171 (2014). disease 2019 (COVID-19) in China. Clin. Inf. Dis. ciaa578 (2020). 38. Tegally, H. et al. Emergence and rapid spread of a new severe acute 11. Williamson, E. et al. Factors associated with COVID-19-related death using respiratory syndrome-related coronavirus 2 (SARS-CoV-2) lineage with OpenSAFELY. Nature 584, 430–436 (2020). multiple spike mutations in South Africa. Preprint at medRxiv https://doi. 12. COVID-19 Signifcantly Impacts Health Services for Noncommunicable Diseases org/10.1101/2020.12.21.20248640 (2020). (World Health Organization, 2020); https://www.who.int/news-room/ 39. Mina, M. J. et al. A global immunological observatory to meet a time of detail/01-06-2020-covid-19-signifcantly-impacts-health-services-for-noncom pandemics. eLife 9, e58989 (2020). municable-diseases 13. Maintaining Essential Health Services: Operational Guidance for the COVID-19 Context (World Health Organization, 2020); https://www.who.int/ Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in publications/i/item/covid-19-operational-guidance-for-maintaining-essent published maps and institutional affiliations. ial-health-services-during-an-outbreak(2020). © The Author(s), under exclusive licence to Springer Nature America, Inc. 2021

Nature Medicine | VOL 27 | March 2021 | 447–453 | www.nature.com/naturemedicine 453 Letters NATURE MEDICInE

Methods health status in these settings47. Although interactions with such infectious diseases Reported SARS-CoV-2 case counts, mortality and testing in SSA as of have been suggested, evidence is limited to date, except for HIV, where effects have 48 December 2020. Variables and data sources for reporting data. Te numbers of been suggested to be minor . We also note that the key concern raised around such reported cases, deaths and tests for the 48 SSA countries studied (Supplementary infections to date is associated with disruption to routine screening (for example, 49 50 51 Table 1) were sourced from the Africa CDC dashboard on 20 December 2020 (and for malaria ), treatment or prevention programs . previously on 23 September and 30 June 2020). Te Africa CDC obtains data from Data on the identified indicators were sourced in May 2020 from the WHO the ofcial Africa CDC Regional Collaborating Centre and member state reports. Global Health Observatory database (https://www.who.int/data/gho), World Bank Diferences in the timing of reporting by member states results in some variation in (https://data.worldbank.org/) and other sources detailed in Supplementary Table 3. the recency of data within the centralized Africa CDC repository, but data should National-level demographic data (population size and age structure) was sourced broadly refect the relative scale of testing and reporting eforts across countries. from the United Nations World Population Prospects and data on subnational 27 For Mauritius (https://covid19.mu/) and Rwanda (https://covid19.who.int/region/ variation in demography was sourced from WorldPop . Household size data was afro/country/rw), reporting to the Africa CDC was confrmed by comparison to defined by the mean number of individuals in a household with at least 1 person country-specifc dashboards. aged >50 years, taken from the most recently available Demographic and Health The countries or member states within SSA in this study follow the United Surveys data (https://dhsprogram.com). All country-level data for all indicators Nations and Africa CDC-listed regions of Southern, Western, Central and Eastern can be found online at the SSA-SARS-CoV-2 tool (https://labmetcalf.shinyapps.io/ Africa (excluding Sudan). From the Northern Africa region, Mauritania is included covid19-burden-africa/). in SSA. Comparisons of national-level estimates sourced from the WHO and other For comparison to non-SSA countries, the number of reported cases in other sources are affected by variation within countries and variation in the uncertainty geographical regions were obtained from the Johns Hopkins University Coronavirus around estimates from different geographical areas. To assess potential differences Resource Center on 23 September 2020 (https://coronavirus.jhu.edu/map.html). in data quality between geographical areas, we compared the year of the most Case fatality ratios (CFRs) were calculated by dividing the number of reported recent data for the variables (Extended Data Fig. 2). The mean (range varied from deaths by the number of reported cases and expressed as a percentage. Positivity 2014.624 to 2014.928 by region) and median year (2016 for all regions) of the most was calculated by dividing the number of reported cases by the number of recent data varied little between regions. To account for the uncertainty associated reported tests. Testing and case rates were calculated per 100,000 population using with the estimates available for a single variable, we also included multiple variables population size estimates for 2020 from the United Nations Population Division per category (for example, demographic and socioeconomic factors, comorbidities, (https://population.un.org/wpp/Download/Standard/Population/). Since reported access to care) to avoid reliance on a single metric. This allowed exploring variation confirmed cases are likely to be an underestimate of the true number of infections, between countries across a broad suite of variables likely to be indicative of the CFRs may be a poor proxy for the IFR, defined as the proportion of infections that different dimensions of risk. result in mortality4. Although including multiple variables that were likely to be correlated (see the principal component analysis (PCA) methods below for further discussion) would Variation in testing and mortality rates. Testing rates among SSA countries varied bias inference of cumulative risk in a statistical framework, we did not attempt to by multiple orders of magnitude as of 30 June and remain highly variable as of 23 quantitatively combine risk across variables for a country, nor project risk based September and 20 December 2020. The number of tests completed per 100,000 on the variables included in this study. Rather, we characterized the magnitude of population ranged from 19.84 in Burundi to 13,508.13 in Mauritius in June 2020; variation among countries for these variables (see Fig. 3 for a subset of the variables 52 from 65.98 in the Democratic Republic of the Congo to 18,321.83 in Mauritius and Fig. 4b for the bivariate risk maps following Chin et al. ) and then explored in September 2020; and from 100.9 in the Democratic Republic of the Congo the range of outcomes that would be expected under scenarios where the IFR to 23695.0 in Mauritius in December 2020 (Extended Data Fig. 1a). Tanzania increases with age at different rates (Fig. 4). (6.50 tests per 100,000 population) has not reported new tests, cases or deaths to the Africa CDC since April 2020. The number of reported infections (that Variable selection and data sources for variables modulating the rate of viral is, positive tests) was strongly correlated with the number of tests completed in spread. In addition to characterizing variation among factors likely to modulate June 2020 (Pearson’s correlation coefficient, r = 0.9667, P < 0.001), September burden, we also synthesized data sources relevant to the rate of viral spread, or 2020 (r = 0.9689, P < 0.001) and December 2020 (r = 0.9750, P < 0.001) (Extended pace, for the SARS-CoV-2 pandemic in SSA. Factors hypothesized to modulate Data Fig. 1b). As of June 2020, no deaths due to SARS-CoV-2 were reported to viral transmission and geographical spread include climatic factors (for example, the Africa CDC for five SSA countries (Eritrea, Lesotho, Namibia, Seychelles, specific humidity), access to prevention measures (for example, handwashing) and Uganda). As of December 2020, still no deaths due to SARS-CoV-2 were reported human mobility (for example, international and domestic travel). Supplementary to the Africa CDC for two of those countries (Eritrea and Seychelles). Among Table 2 outlines the dimensions of risk selected and references the previous studies countries with at least 1 reported death, the CFR varied from 0.22% in Rwanda relevant to the selection of these factors. 53 to 8.54% in Chad in June 2020; from 0.21% in Burundi to 6.96% in Chad in Climate data were sourced from the global, gridded ERA5 dataset where September 2020; and from 0.26% in Burundi to 5.40% in Chad in December model data were combined with global observation data (see Methods for 2020 (Extended Data Fig. 1c). Limitations in the ascertainment of infection rates climate-driven modeling of SARS-CoV-2 section for further details). and the rarity of reported deaths (for example, the median number of reported International flight data were obtained from a custom report from OAG deaths per SSA country was 25.5 as of June 2020, 71.0 as of September 2020 and Aviation Worldwide (UK) and included the departure location, arrival airport, date 101.0 as of December 2020), indicate that the data are insufficient to determine of travel and number of passenger seats for flights arriving to 113 international country-specific IFRs and IFR by age profiles for most countries. As a result, global airports in SSA (see International air travel to SSA section). IFR by age estimates was used for the subsequent analyses in this study. As an estimate of connectivity within subregions of countries, the population-weighted mean travel time to the nearest city with a population greater Synthesizing factors that increase or decrease SARS-CoV-2 epidemic risk in than 50,000 was determined; details are provided in the section on Subnational SSA. Variable selection and data sources for variables associated with an increased connectivity among countries in SSA. To obtain a set of measures that broadly probability of severe clinical outcomes for an infection. To characterize epidemic represent connectivity within different countries in the region, friction surfaces risk, defned as potential SARS-CoV-2 related morbidity and mortality, we frst from Weiss et al.26 were used to obtain estimates of the connectivity between synthesized factors hypothesized to infuence risk in SSA settings (Supplementary different administrative level 2 units within each country. Details of this, alongside Table 2). Early during the pandemic, evidence suggested that age was an important the metapopulation model framework used to simulate viral spread with variation risk factor associated with morbidity and mortality associated with SARS-CoV-2 in connectivity are found in the Subnational connectivity section. infection40, a pattern subsequently confrmed across settings2,11,41. Associations Figure 3 shows variation among SSA countries for four of the variables and between SARS-CoV-2 mortality and comorbidities including hypertension, Extended Data Fig. 3 links to visualizations of variation for all variables. Figure diabetes and cardiovascular disease emerged early40 and have been observed across 4 shows variation for a subset of the comorbidity and access to care indicators settings, with further growing evidence for associations with obesity11,42, severe as a heatmap and Extended Data Fig. 4 shows variation for all the variables (also asthma11 and the respiratory efects of pollution43. Specifc to Africa, vulnerability available at https://labmetcalf.shinyapps.io/covid19-burden-africa/). scores based on these hypothesized associations or combinations of risks factors have been developed (for example, refs. 44,45). PCA of variables considered. Selection of data and variables. Te 29 national-level Many possible sources of bias complicate interpretation of these associations46; variables from Supplementary Table 3 were selected for the PCA. We conducted while they provide a useful baseline, inference is also likely to change as the further PCA on the subset of 8 indicators related to access to healthcare (category pandemic advances. To reflect this, our analysis combined a number of high-level E) and the 14 national indicators variables related to comorbidities (category B). variables likely to broadly encompass these putative risk factors (for example, We excluded disaggregated subnational spatial variation data (variables A2, C1, noncommunicable disease-related mortality and healthy life expectancy) E2 and category F), disaggregated or redundant variables derived from variables with more specific measures encompassed in evidence to date (for example, already included (variables A4 and D2) and disaggregated age-specific disease data prevalence of diabetes, obesity and respiratory illness, such as chronic obstructive from the Institute for Health Metrics and Evaluation (IHME) global burden of pulmonary disease). We also included measures relating to infectious diseases, disease study (variables B2, B4 and B13) from the PCA analysis. COVID-19 tests undernourishment and anemia given their interaction and effects in determining per 100,000 population (variable D4; Supplementary Table 1), per capita gross

Nature Medicine | www.nature.com/naturemedicine NATURE MEDICInE Letters domestic product (GDP) (variable A8) and the Gini index of wealth inequality mortality but depends on the assumption that mortality patterns in SSA are similar (variable A9) were used to visualize patterns among SSA countries. to those from where the IFR estimates were sourced (France, China and ). In some cases, data were missing for a country for an indicator; in these cases, Comorbidities have been shown to be an important determinant of the severity missing data were replaced with a zero value. This is a conservative approach since of infection outcomes (that is, IFR). To assess the relative risk of comorbidities zero values (that is, outside the range of typical values seen in the data) inflate across age in SSA, estimates of comorbidity severity by age (in terms of annual the total variance in the dataset and thus, if anything, deflate the percentage of deaths attributable) were obtained from the IHME Global Burden of Disease the variance explained by the PCA. Therefore, this approach avoids mistakenly (GBD) study in 2017 (ref. 59). Data were accessed through the GBD results tool attributing predictive value to principal components due to incomplete data. See for cardiovascular disease, chronic respiratory disease (not including asthma) Supplementary Table 3 for data sources for each variable. and diabetes, reflecting three categories of comorbidity with demonstrated associations with risk (Supplementary Table 2). We assumed that higher mortality PCA. The PCA was conducted on each of the three subsets described above using rates due to these noncommunicable diseases, especially among younger age the scikit-learn library54. To avoid biasing the PCA due to large differences in groups, is indicative of increased severity and lesser access to sufficient care for magnitude and scale, each feature was centered around the mean and scaled to these diseases, suggesting an elevated risk for their interaction with SARS-CoV-2 unit variance before the analysis. Briefly, PCA applies a linear transformation to as comorbidities. While there are uncertainties in these data, they provide the a set of n features to output a set of n orthogonal principal components that are best estimates of age-specific risks and have been used previously to estimate uncorrelated and each explain a percentage of the total variance in the dataset55. populations at risk20. A link to the code for this analysis is available at https://labmetcalf.shinyapps.io/ The comorbidity by age curves for SSA countries were compared to those for covid19-burden-africa/. the three countries from which SARS-CoV-2 IFR by age estimates were sourced. The principal components were then analyzed for the percentage of variance Attributable mortality due to all three noncommunicable disease categories was explained and compared to: (1) the number of COVID-19 tests per 100,000 higher at age 50 in all 48 SSA countries when compared to estimates from France population as of the end of June 2020 (Supplementary Table 1); (2) the per capita and Italy and for 42 of 48 SSA countries when compared to China (Extended GDP; and (3) the Gini index of wealth inequality. For the Gini index, estimates Data Fig. 6). from 2008 to 2018 were available for 45 of the 48 countries (no Gini index data Given the potential for populations in SSA to experience a differing burden were available for Eritrea, Equatorial Guinea and Somalia). of SARS-CoV-2 due to their increased severity of comorbidities in younger age The first 2 principal components from the analysis of 29 variables explain 32.6 groups, we explored the effects of shifting IFRs estimated by the generalized and 13.1% the total variance, respectively, in the dataset. Countries with higher additive model of IFR estimates from France, Italy and China younger by 2, 5 and numbers of completed SARS-CoV-2 tests reported tended to associate with an 10 years (Fig. 3). increase in principal component 1 (r = 0.67, P = 1.1 × 10−7; Extended Data Fig. 5a). Similarly, countries with a high GDP seemed to associate with an increase International air travel to SSA. The number of passenger seats on flights arriving in principal component 1 (r = 0.80, P = 6.02 × 10−12; Extended Data Fig. 5b). In to international airports were grouped by country and month for January to April contrast, countries with greater wealth inequality (as measured by the Gini index) 2020 (Supplementary Table 5), the months when the introduction of SARS-CoV-2 were associated with a decrease in principal component 2 (r = −0.42, P = 0.0042; to SSA countries was likely to have first occurred. The first confirmed case Extended Data Fig. 5c). Despite these correlations, a relatively low percentage of reported from an SSA country, according to the Johns Hopkins Coronavirus variance was explained by each principal component: for the 29 variables, 13 of the Research Center was in Nigeria on 28 February 2020. By 31 March 2020, 43 of 29 principal components were required to explain 90% of the variance (Extended 48 SSA countries had reported SARS-CoV-2 infections and international travel Data Fig. 5d). When only the access to care subset of variables is considered, the was largely restricted by April. Lesotho was the last SSA country to report a first 2 principal components explain 50.7 and 19.1% of the variance, respectively, confirmed SARS-CoV-2 infection (on 13 May 2020); however, given the difficulties and 5 of 8 principal components are required to explain 90% of the variance. When in surveillance, the first reported detections were likely delayed relative to the only the comorbidities subset is considered, the first two principal components first importations of the virus. The probability of importation of the virus is explain 27.9 and 17.8% of the variance, respectively, and 9 of 14 principal defined by the number of travelers from each source location, each date and the components are required to explain 90% of the variance (Extended Data Fig. 5d). probability that a traveler from that source location on that date was infectious. These data suggest that intercountry variation in this dataset is not easily Due to limitations in surveillance, especially early in the SARS-CoV-2 pandemic, explained by a small number of variables. Moreover, although correlations exist empirical data on infection rates among travelers were largely lacking. To account between principal components and high-level explanatory variables (testing capacity, for differences in the status of the SARS-CoV-2 pandemic across source locations wealth), their magnitude is modest. These results highlight that dimensionality and thus differences in the importation risk for travelers from those locations, we reduction is unlikely to be an effective analysis strategy for the variables considered coarsely stratified travelers arriving each day into 4 categories based on the status in this study. Despite this overall finding, the PCA on the access to care subset of of their source countries: (1) travelers from countries with zero reported cases variables highlights that the variance in these variables is more easily explained (that is, although undetected transmission was possibly occurring, SARS-CoV-2 by a small number of principal components and hence may be more amenable to had not yet been confirmed in the source country by that date); (2) those traveling dimensionality reduction. This finding is unsurprising since, for example, the number from countries with more than 1 reported case (that is, SARS-CoV-2 had been of hospital beds per 100,000 population is likely to be directly related to the number confirmed to be present in that source country by that date); (3) those traveling of hospitals per 100,000 population (r = 0.60, P = 5.7 ×10 −6 for SSA). In contrast, from countries with more than 100 reported cases (indicating community for comorbidities, the relationship between different variables is less clear. Given transmission was likely beginning); and (4) those traveling from countries with the low percentages of variation captured by each principal component, and the more than 1,000 reported cases (indicating widespread transmission). high variability between different types of variables, these results motivate a holistic To determine reported case counts at source locations for travelers, no cases approach to using these data for assessing relative SARS-CoV-2 risk across SSA. were reported outside China until 13 January 2020 (the date of the first reported case in ). Over 13 January to 21 January, cases were then reported in Evaluating the burden emerging from the severity of infection outcome. Japan, , Taiwan, Hong Kong and the United States (https://covid19. Data sourcing: empirical estimates of IFR. Estimates of the IFR that account for who.int/). Subsequently, counts per country were tabulated daily by the Johns asymptomatic cases, underreporting and delays in reporting are few; however, it Hopkins Coronavirus Resource Center60 beginning on 22 January (https:// is evident that the IFR increases substantially with age56. We used age-stratifed coronavirus.jhu.edu/map.html); we used the data from 22 January onwards and the estimates of IFR from three studies (two published2,4 and one preprint3) that WHO reports before 22 January. accounted for these factors in their estimation (Supplementary Table 4). The number of travelers within each category arriving per month is shown in To apply these estimates to other age-stratified data with different bin ranges Supplementary Table 5. This approach makes the conservative assumption that and generate continuous predictions of IFR with age, we fitted the relationship the probability a traveler is infected reflects the general countrywide infection between the midpoint of the age bracket and the IFR estimate using a generalized rate of the source country at the time of travel (that is, travelers are not more additive model using the mgcv package v1.8-33 (ref. 57) in R v.4.0.2 (ref. 58). We likely to be exposed than non-travelers in that source location) and does not used a beta distribution as the link function for the IFR estimates (data distributed account for complex travel itineraries (that is, a traveler from a high-risk source on [0, 1]). For the upper age bracket (80+ years), we took the upper range to be 100 location transiting through a low-risk source location would be grouped with years and the midpoint to be 90. other travelers from the low-risk source location). Consequently, the risk for viral We assumed a given level of cumulative infection (20% in each age class, that is, importation is likely systematically underestimated. However, since the relative a constant rate of infection among age classes) and then applied IFRs by age to the risk for viral importation will still scale with the number of travelers, comparisons population structure of each country to generate estimates of burden. Age structure among SSA countries can be informative (for example, SSA countries with more estimates were taken from the United Nations World Population Prospects travelers from countries with confirmed SARS-CoV-2 transmission are at higher (Supplementary Table 3) country-level estimates of population in 1-year age groups risk for viral importation). (0–100 years of age) to generate estimates of burden. Subnational connectivity among countries in SSA. Indicators of subnational Comorbidities over age from the IHME. Applying these IFR estimates to the connectivity. To allow comparison of the relative connectivity across countries, demographic structure of SSA countries provides a baseline expectation for we used the friction surface estimates provided by Weiss et al.26 as a relative

Nature Medicine | www.nature.com/naturemedicine Letters NATURE MEDICInE measure of the rate of human movement between subregions of a country. For The U-shaped pattern diminishes as the rate of waning of immunity increases and connectivity within subregions of a country (for example, transport from a city to is replaced by a monotonic negative relationship. With sufficiently rapid waning of the rural periphery), we used as an indicator the population-weighted mean travel immunity, local extinction ceases to occur in the absence of control efforts. time to the nearest urban center (that is, population density >1,500 per square The impact of the pattern of travel between centroids is echoed by the pattern of kilometer or a density of built-up areas >50% coincident with population >50,000) travel within administrative districts: countries where the pathogen does not reach within administrative 2 units61. For some countries, estimates at administrative 2 a large fraction of the administrative 2 units within the country in five years are also units were unavailable (Comoros, Cape Verde, Lesotho, Mauritius, and those where within-administrative-unit travel is low (Extended Data Fig. 7, right). Seychelles); estimates at the administrative-1 unit level were used for these cases These simulations provide a window into qualitative patterns expected for (these were all island nations, with the exception of Lesotho). subnational spread of the pandemic virus but there is no clear way of calibrating the absolute rate of travel between regions of relevance for SARS-CoV-2; this is further Metapopulation model methods. Once SARS-CoV-2 has been introduced into a complicated by the remaining uncertainties around rates of waning of immunity. country, the degree of spread of the infection within the country is governed by Thus, the time scales of these simulations should be considered in relative, rather than subnational mobility: the pathogen is more likely to be introduced into a location absolute terms. Variation in lockdown effectiveness, or other changes in mobility for where individuals arrive more frequently than one where incoming travelers a given country, may also compromise relative comparisons as might large volumes are less frequent. Large-scale consistent measures of mobility are rare. However, of land border crossings in some settings, which we have not accounted for in this recently, estimates of accessibility have been produced at a global scale26. Although study. Variability in testing and case reporting complicates clarifying this (Extended this is unlikely to perfectly reflect mobility within countries, especially since Data Fig. 7, bottom left and bottom right, respectively) but we have highlighted interventions and travel restrictions are put in place, it provides a starting point countries with less connectivity (that is, less synchronous outbreaks expected) relative for evaluating the role of human mobility in shaping the outbreak pace across to the median among SSA countries and with older populations (that is, a greater SSA. We used the inverse of a measure of the cost of travel between the centroids proportion in higher-risk age groups) (Extended Data Fig. 8). of administrative level 2 spatial units to describe mobility between locations The University of Oxford’s Blavatnik School of Government generated (estimated by applying the costDistance function in the gdistance package v1.3-6 composite scores of government response, interventions for containment and in R to the friction surfaces supplied in Weiss et al.26). With this, we developed a economic support provided, with each scored from 0 to 100 (Coronavirus metapopulation model for each country to develop an overview of the possible Government Response Tracker; https://www.bsg.ox.ac.uk/research/ range of trajectories of unchecked spread of SARS-CoV-2. research-projects/coronavirus-government-response-tracker). These data were We assumed that the pathogen first arrives in each country in the administrative compared with the day on which ten cases were exceeded in a country according to 2 level unit with the largest population (for example, the largest city) and the the Johns Hopkins dashboard data (Johns Hopkins Coronavirus Resource Center; population in each administrative 2 level (of size Nj) is entirely susceptible at https://coronavirus.jhu.edu/map.html). the time of arrival. We then tracked the spread within and between each of the While faster waning of immunity will act to increase the rate of spread of the administrative 2 level units of each country. Within each administrative 2 level unit, infection, resulting in a higher proportion infected after one year, control efforts dynamics are governed by a discrete time susceptible (S), infected (I) and recovered will generally act to slow the rate of spread of the infection (Extended Data Fig. 9). (R) model with a time step of approximately one week, which is broadly consistent Since different countries are likely to have differently effective control efforts with the serial interval of SARS-CoV-2. Within the spatial unit indexed j, with total (Extended Data Fig. 9), this precludes making country-specific predictions as to size Nj, the number of infected individuals in the next time step is defined by: the relative impact of control efforts on delay. α Ij;t 1 βIj;t Sj;t =Nj ιj;t þ ¼ þ Modeling epidemic trajectories in scenarios where transmission rate depends on climate. Climate data sourcing: variation in humidity in SSA. Specifc humidity where β captures the magnitude of transmission over the course of one discrete data for selected urban centers comes from the ERA5 using an average climatology time step; since the discrete time step chosen is set to approximate the serial (1981–2017)53; we did not consider year-to-year climate variations. Selected cities interval of the virus, this will reflect the R0 of SARS-CoV-2, and is thus set to (n 56) were chosen to represent the major urban areas in SSA. Te largest city in 62 = 2.5; the exponent α = 0.97 is used to capture the effects of discretization and Ij,t each SSA country was included as well as any additional cities that were among the captures the introduction of new infections into site j at time t. Susceptible and 25 largest cities or busiest airports in SSA. recovered individuals are updated according to: Methods for climate-driven modeling of SARS-CoV-2. We used a climate-driven Sj;t 1 Sj;t wRj;t Ij;t 1 b þ ¼ þ � þ þ susceptible-infected-recovered-susceptible model to estimate epidemic trajectories Rj;t 1 1 w Rj;t Ij;t þ ¼ð � Þ þ (that is, the time of peak incidence) in different cities in 2020, assuming no control

measures were in place or a 10 or 20% reduction in R0 beginning 2 weeks after the where b reflects the introduction of new susceptible individuals resulting from the 25,63 birth rate, set to reflect the most recent estimates for that country from the World total reported cases for a country exceeded 10 cases . The model is given by: Bank Data (https://data.worldbank.org/indicator/SP.DYN.CBRT.IN), and w reflects dS N S L β t IS � � ð Þ the rate of waning of immunity. The population is initiated with Sj,1 = NjRj,1 = 0, and dt ¼ L � N Ij,1 = 0 except for the spatial unit corresponding to the largest population size Nj for each country since this is assumed to be the location of introduction; for this spatial dI β t IS I ð Þ unit, we set Ij,1 = 1. dt ¼ N � D We made the simplifying assumption that mobility linking locations i and j, denoted as ci,j, scales with the inverse of the cost of travel between sites i and where S is the susceptible population, I is the infected population and N is the total j evaluated according to the friction surface provided in Weiss et al.26. The population. D is the mean infectious period, set at 5 d following ref. 25. introduction of an infected individual into location j is then defined by a draw To investigate the effects on epidemic trajectories of a climate dependency of from a Bernouilli distribution following: SARS-CoV-2 on cities with the climate patterns of the selected cities in SSA, we used parameters from the most climate-dependent scenario in ref. 25, based on L the endemic betacoronavirus HKU1 in the United States. In this scenario L, the ι Bernouilli 1 exp c I =N j;t i;j i;t i duration of immunity, was 66.25 weeks (that is, 1 year and such that waning  À À 1 ! ! > X immunity did not affect the timing of the epidemic peak). We initially selected where L is the total number of administrative 2 units in that country and the rate a range where R0 declined from R0max = 2.5 to R0min = 1.5 (that is, transmission of introduction is the product of connectivity between the focal location and each declined 40% at high humidity) since this exceeds the range observed for other location multiplied by the proportion of population in each other location influenza and other coronaviruses for which data are available (from the United that is infected. States). R0max = 2.5 was chosen because 2.5 is often cited as the approximate R0 Some countries show rapid spread between administrative units within for SARS-CoV-2. Thus, we initially assumed that the climate dependence of the country (for example, a country with parameters that broadly reflect those SARS-CoV-2 in SSA would not greatly exceed that of other known coronaviruses available for Malawi; Extended Data Fig. 7), while in others (for example, from the US context. Then, we explored the effects of different degrees of climate reflecting Madagascar), connectivity may be so low that the outbreak may be dependency (that is, wider ranges between R0max = 2.5 to R0min = 1.5 and scenarios over in the administrative unit of the largest size (where it was introduced) before where R0min approached 1) (Extended Data Fig. 10). introductions successfully reach other poorly connected administrative units. Transmission is governed by β(t), which is related to the basic reproduction Where duration of immunity is sufficiently long, the result may be a hump-shaped number R0 by R0(t) = β(t)D. The basic reproduction number varies based on relationship between the proportion of the population that is infected after five climate and is related to specific humidity according to the equation: years and the time to the first local extinction of the pathogen (Extended Data R0 exp a ´ q t log R0max R0min R0min Fig. 7, top right). In countries with lower connectivity (for example, resembling ¼ ½ ð Þþ ð � Þ þ Madagascar), local outbreaks can go extinct rapidly before traveling very far; in where q(t) is specific humidity53 and a is set at −227.5 based on estimated HKU1 other countries (for example, resembling Gabon), the pathogen goes extinct rapidly parameters25. We assumed the time of introduction for cities to be the date at which because it travels rapidly and rapidly depletes susceptible individuals everywhere. the total reported cases for a country exceeded 10 cases.

Nature Medicine | www.nature.com/naturemedicine NATURE MEDICInE Letters

Sensitivity analysis. Selecting an R0min value of 1, such that epidemic growth 55. Jollife, I. T. & Cadima, J. Principal component analysis: a review stops at high humidities, is likely implausible since simulations indicated and recent developments. Philos. Trans. A Math. Phys. Eng. Sci. 374, no outbreaks would occur in cities such as Antananarivo (countered by the 20150202 (2016). observation that SARS-CoV-2 outbreaks did in fact occur) (Extended Data Fig. 56. Meyerowitz-Katz, G. & Merone, L. A systematic review and meta-analysis of 10b; see Supplementary Table 1 for the reported case counts at the country level). published research data on COVID-19 infection fatality rates. Int. J. Infect.

Expanding the range between R0min and R0max by increasing R0max resulted in Dis. 101, 138–148 (2020). epidemic peaks being reached earlier after outbreak onset but did not increase the 57. Wood, S. N. Generalized Additive Models: An Introduction with R difference in timing between cities with different climates (Extended Data Fig. 10c; (CRC Press, 2017). for example, the difference in timing between peaks in Windhoek and Lomé is 58. R Core Team. R: A Language and Environment for Statistical Computing. similar in 10a and 10c). Finally, we explored scenarios where the R0min was between https://www.R-project.org/ (R Foundation for Statistical Computing, 2020). 1.0 and 1.5. When R0min > 1.1, epidemic peaks were seen in each SSA city with 59. Global Burden of Disease Study 2017 (GBD 2017) Population Estimates the difference in timing of the peak growing larger when smaller values of R0min 1950–2017 (Institute for Health Metrics and Evaluation, 2019); http://ghdx. were selected (Extended Data Fig. 10d). However, the difference in timing, even healthdata.org/record/ihme-data/gbd-2017-population-estimates-1950-2017 when small values of R0min were selected, was a maximum of 25 weeks and rapidly 60. Dong, E., Du, H. & Gardner, L. An interactive web-based dashboard to track reduced to only a few weeks when R0min approached 1.5. COVID-19 in real time. Lancet Infect. Dis. 20, 533–534 (2020). 61. Linard, C., Gilbert, M., Snow, R. W., Noor, A. M. & Tatem, A. J. Population Reporting Summary. Further information on research design is available in the distribution, settlement patterns and accessibility across Africa in 2010. PLoS Nature Research Reporting Summary linked to this . ONE 7, e31743 (2012). 62. Bjørnstad, O. N., Finkenstäd, B. F. & Grenfell, B. T. Dynamics of measles Data availability epidemics: estimating scaling of transmission rates using a time series SIR All data have been deposited into a publicly available GitHub repository at https:// model. Ecol. Monogr. 72, 169–184 (2002). github.com/labmetcalf/SSA-SARS-CoV-2. 63. Shaman, J., Pitzer, V. E., Viboud, C., Grenfell, B. T. & Lipsitch, M. Absolute humidity and the seasonal onset of infuenza in the continental United States. PLoS Biol. 8, e1000316 (2010). Code availability 64. COVID-19 Community Mobility Reports. https://www.google.com/covid19/ All code has been deposited into the publicly available GitHub repository at https:// mobility/ (Google, 2020). github.com/labmetcalf/SSA-SARS-CoV-2. 65. Measles and Rubella Surveillance Data. http://158.232.12.119/immunization/ monitoring_surveillance/burden/vpd/surveillance_type/active/measles_ References monthlydata/en/ (World Health Organization, 2020). 40. Novel Coronavirus Pneumonia Emergency Response Epidemiology Team. 66. Nimpa, M. M. et al. Measles outbreak in 2018–2019, Madagascar: Te epidemiological characteristics of an outbreak of 2019 novel coronavirus epidemiology and public health implications. Pan Afr. Med. J. 35, 84 (2020). diseases (COVID-19): China, 2020. China CDC Wkly. 2, 113–122 (2020). 41. Bialek, S. et al. Severe outcomes among patients with coronavirus disease Acknowledgements 2019 (COVID-19): United States, February 12–March 16, 2020. MMWR R.E.B. is supported by the Cooperative Institute for Modeling the Earth System. Morb. Mortal. Wkly. Rep. 69, 343–346 (2020). A.A. acknowledges support from the National Institutes of Health Medical Scientist 42. Simmonet, A. et al. High prevalence of obesity in severe acute respiratory Training Program no. 1T32GM136577. A.J.T. is funded by the Bill & Melinda Gates syndrome coronavirus-2 (SARS-CoV-2) requiring invasive mechanical Foundation (nos. OPP1182425, OPP1134076 and INV-002697). M.B. is funded by ventilation. Obesity (Silver Spring) 28, 1195–1199 (2020). Nederlandse Organisatie voor Wetenschappelijk Onderzoek (Dutch Rsearch Council) 43. Conticini, E., Frediani, B. & Caro, D. Can atmospheric pollution be Rubicon grant no. 019.192EN.017. W.D.-G. is supported by ESRC SCDTP grant considered a co-factor in extremely high level of SARS-CoV-2 lethality in number ES/P000673/1. We thank the Center for Health and Wellbeing, Princeton Northern Italy? Environ. Pollut. 261, 114465 (2020). University for support. 44. Mapping Risk Factors for the Spread of COVID-19 in Africa https:// africacenter.org/spotlight/mapping-risk-factors-spread-covid-19-africa/ (Africa Center for Strategic Studies, 2020). Author contributions 45. Te Africa COVID Community Vulnerability Index (CCVI). An Index that B.L.R., A.A., R.E.B., M.B., W.D.-G, K.M., I.F.M., N.V.M., A.R., M.R., J.R., T.R., F.R., W.Y., Assesses Health, Economic, and Social Impacts of COVID-19 in Africa. B.T.G., A.J.T. and C.J.E.M. conceived the analysis. B.L.R., M.R., M.B., W.D.-G. and W.Y. https://precisionforcovid.org/africa (Precision for COVID, 2020). are responsible for data curation. B.L.R., A.A., M.B., M.R., R.E.B. and C.J.E.M. performed 46. Grifth, G. J. et al. Collider bias undermines our understanding of the analyses. B.L.R., M.R., M.B., R.E.B., C.J.E.M. and B.T.G. are responsible for the COVID-19 disease risk and severity. Nat. Commun. 11, 5749 (2020). methodology. B.L.R., M.R., M.B., R.E.B. and W.Y. are responsible for the software and the 47. Shankar, A. H. Nutritional modulation of malaria morbidity and mortality. Shiny app online tool. B.L.R., M.R., M.B., R.E.B. and W.Y. led the visualization of data. J. Infect. Dis. 182, S37–S53 (2000). B.L.R. and C.J.E.M. wrote the initial draft of the manuscript. B.L.R., A.A., R.E.B., M.B., 48. Cooper, T. J., Woodward, B. L., Alom, S. & Harky, A. Coronavirus disease W.D.-G., K.M., I.F.M., N.V.M., A.R., M.R., J.R., T.R., F.R., W.Y., B.T.G., A.J.T. and C.J.E.M. contributed to reviewing and editing the manuscript. 2019 (COVID‐19) outcomes in HIV/AIDS patients: a systematic review. HIV Med. 21, 567–577 (2020). 49. Chanda-Kapata, P., Kapata, N. & Zumla, A. COVID-19 and malaria: a Competing interests symptom screening challenge for malaria endemic countries. Int. J. Infect. Dis. The authors declare no competing interests. 94, 151–153 (2020). 50. Lesosky, M. & Myer, L. Modelling the impact of COVID-19 on HIV. Lancet HIV 7, e596–e598 (2020). Additional information 51. Sherrard-Smith, E. et al. Te potential public health consequences of Extended data is available for this paper at https://doi.org/10.1038/s41591-021-01234-8. COVID-19 on malaria in Africa. Nat. Med. 26, 1411–1416 (2020). Supplementary information The online version contains supplementary material 52. Chin, T. et al. US-county level variation in intersecting individual, household available at https://doi.org/10.1038/s41591-021-01234-8. and community characteristics relevant to COVID-19 and planning an Correspondence and requests for materials should be addressed to B.L.R. equitable response: a cross-sectional analysis. BMJ Open 10, e039886 (2020). 53. Hersbach, H. et al. Te ERA5 global reanalysis. Q. J. R. Meteorol. Soc. 146, Peer review information Jennifer Sargent was the primary editor on this article and 1999–2049 (2020). managed its editorial process and peer review in collaboration with the rest of the 54. Pedregosa, F. et al. Scikit-learn: machine learning in Python. J. Mach. Learn. editorial team. Res. 12, 2825–2830 (2011). Reprints and permissions information is available at www.nature.com/reprints.

Nature Medicine | www.nature.com/naturemedicine Letters NATURE MEDICInE

Extended Data Fig. 1 | Variation between SSA countries in testing and reporting rates. a, Reported number of tests completed per country as of December 20, 2020. b, Number of infections (I) per reported number of tests (T); line shows linear least squares regression: I = 1.422×10-1×T – 1.912×104 (df = 46, adjusted R2 = 0.9496, Pearson’s correlation coefficient, r = 0.9750, p < 0.001). c, Reported infections and deaths for sub-Saharan African countries with case fatality ratios (CFRs) shown as diagonal lines.

Nature Medicine | www.nature.com/naturemedicine NATURE MEDICInE Letters

Extended Data Fig. 2 | Year of most recent data available for variables compared between global regions. Dotted vertical line shows regional median; solid vertical line shows regional mean. Note that most data comes from 2015-2019 (median = 2016, mean = 2014.62-2014.93).

Nature Medicine | www.nature.com/naturemedicine Letters NATURE MEDICInE

Extended Data Fig. 3 | Variation among sub-Saharan African countries in determinants of SARS-CoV-2 risk by variable. A subset of variables is shown in Fig. 3a–d in the main text. Non-communicable disease (NCD) overall mortality per 100,000 population (age standardized) is shown here as an exemplar. The remaining variables are shown online: SSA-SARS-CoV-2-tool (https://labmetcalf.shinyapps.io/covid19-burden-africa/).

Nature Medicine | www.nature.com/naturemedicine NATURE MEDICInE Letters

Extended Data Fig. 4 | Variation among sub-Saharan African countries in determinants of SARS-CoV-2 mortality risk by category. A subset of variables is shown in Fig. 4d,e in the main text. The remaining variables are shown online: SSA-SARS-CoV-2-tool (https://labmetcalf.shinyapps.io/ covid19-burden-africa/). a, Select national level indicators; estimates of increased comorbidity burden (for example, higher prevalence of raised blood pressure) shown with darker red for higher risk quartiles. b, Select national level indicators; estimates of reduced access to care (for example, fewer hospitals) shown with darker red for higher risk quartiles. Countries missing data for an indicator (NA) are shown in gray. For comparison between countries, estimates are age-standardized where applicable (see Supplementary Table 3 for details).

Nature Medicine | www.nature.com/naturemedicine Letters NATURE MEDICInE

Extended Data Fig. 5 | Principal Component Analysis of all variables and category specific subsets of variables. a, Principal Component 1 and 2, countries colored by Log10 scaled tests per 100,000 population (as of June 30, 2020). b, Principal Component 1 and 2, countries colored by Log10 scaled GDP per capita. c, Principal Component 1 and 2, countries colored by the GINI index (a measure of wealth disparity). d, Scree plot showing the cumulative proportion of variance explained by principal component for analysis done using all variables (blue, 29 variables), comorbidity indicators (green, 14 variables, Section B in Supplementary Table 3)), and access to care indicators (orange, 8 variables, Section E in Supplementary Table 3).

Nature Medicine | www.nature.com/naturemedicine NATURE MEDICInE Letters

Extended Data Fig. 6 | Comorbidity burden by age in sub-Saharan Africa. Estimated mortality per age group for sub-Saharan African countries (gray lines) compared to China, France, and Italy (the countries from which estimates of SARS-CoV-2 infection fatality ratios (IFRs) by age are available) for three NCD categories (cardiovascular diseases, chronic respiratory diseases excluding asthma, and diabetes).

Nature Medicine | www.nature.com/naturemedicine Letters NATURE MEDICInE

Extended Data Fig. 7 | Pace of the outbreak and cases and testing vs. the pace of the outbreak. Top: Each grey line on the left-hand panels indicates the total infected across all administrative units in a metapopulation simulation with parameters reflecting the country indicated by the plot title, assuming interventions are constant, and that immunity does not wane. Simulations with parameters reflecting three representative countries are shown, ranked from higher connectivity (Malawi-like) to lower connectivity (Madagascar-like). The top right-hand plot shows where more rapid disappearances of the outbreak locally are expected (y axis shows time to first extinction) and where a higher proportion of the countries’ population is reached during simulation (x axis shows proportion of population infected by 1 year); grey horizontal bars indicate quartiles across 100 simulations. We note that a shorter duration of immunity will reduce the probability of extinction within an admin-2 (simulations shown do not include waning). The lower right-hand panel shows the fraction of administrative units unreached against the travel time in hours to the nearest city of 50,000 or more people; grey horizontal bars again reflect quartiles across 100 simulations. Bottom: The total number of confirmed cases reported by country (x axis, left, as reported for June 28th by Africa CDC) and the test positivity (x axis, right, defined as the total number of confirmed cases divided by the number of tests run, as reported by Africa CDC) compared with the proportion of the population estimated to be infected after one year using the metapopulation simulation described in the methods, assuming no waning of immunity (Pearson’s correlation coefficients, respectively, r = −0.04, p > 0.5, df=41; r = 0.02, p > 0.5, df = 41).

Nature Medicine | www.nature.com/naturemedicine NATURE MEDICInE Letters

Extended Data Fig. 8 | Bivariate example of expected pace versus expected burden at the national level in SARS-CoV-2 outbreaks in sub-Saharan Africa. Countries are colored by with respect to indicators of their expected epidemic pace (using as an example subnational connectivity in terms of travel time to nearest city) and potential burden (using as an example the proportion of the population over age 50). a, In pink, countries with less connectivity (that is, less synchronous outbreaks) relative to the median among SSA countries; in blue, countries with more connectivity; darker colors show countries with older populations (that is, a greater proportion in higher risk age groups). b, Dotted lines show the median; in the upper right, in dark pink, countries are highlighted due to their increased potential risk for an outbreak to be prolonged (see metapopulation model methods) and high burden (see burden estimation methods).

Nature Medicine | www.nature.com/naturemedicine Letters NATURE MEDICInE

Extended Data Fig. 9 | Impact of waning of immunity and the introduction of control efforts on spatial spread. a, Impact of waning of immunity and the introduction of control efforts on spatial spread. The left panel indicates the proportion of the population infected after one year in the absence (x-axis) or presence (y-axis) of waning of immunity (duration of immunity taken to be ~40 weeks, that is, w=1/40, reflecting estimates for other coronaviruses HCoV-OC43 and HCoV-HKU1) across countries in SSA; grey horizontal lines indicate quartiles across 100 simulations. All points above the 0,1 line indicate that waning of immunity accelerates spatial spread. The central panel indicates the proportion of the population infected after one year in the absence (x axis) or presence (y axis) of control efforts with 12 weeks of a 20% reduction in transmission as an exemplar. All points below the 0,1 line indicate a lower proportion infected as a result of control efforts. All points above the 0,1 line in the right panel indicate more weeks until the first extinction in the presence of NPIs. Note that a duration of immunity of less than 40 weeks yields no local extinction. b, Time course of the range of policies deployed across different countries. A composite score of government response (left), interventions for containment (middle) and economic support provided (right) each scored from 0-100, provided by the University of Oxford Blavatnik School of Government; showing SSA countries (black lines) relative to other countries (grey lines). c, Comparison of policies implemented in SSA and google derived measures of mobility. The black line indicates a score of the magnitude of policies directed towards health containment for each country (plot title) on a scale from 0-100 with other SSA countries for which data on mobility was available (n = 24 of 48) shown for comparison in grey; the red line indicates the percent reductions in mobility to work relative to baseline64 for that country (similar patterns seen for other mobility measures). The vertical blue line shows the day on which 10 cases were exceeded based on the Johns Hopkins dashboard data. d, Comparison of reductions in transmission with another directly transmitted infection. Monthly measles incidence (y-axis) between 2011 and 201965 is shown in gray, and the first 6 months of 2020 (months on the x-axis) shown in red for countries for which data is available in SSA (n = 34 of 48). China and Germany (which have been relatively successful in controlling the virus) shown for comparison at the bottom right. Although multi-annual features might drive measles incidence (for example, dynamics in Madagascar are largely dominated by a honeymoon outbreak that occurred in 2018-201966) for countries that slowed the SARS-CoV-2 pandemic, signatures of reduction in measles can be identified (for example, Germany and China; similar patterns are seen in Viet Nam).

Nature Medicine | www.nature.com/naturemedicine NATURE MEDICInE Letters

Extended Data Fig. 10 | Transmission climate-dependency and sensitivity to R0max and R0min value selection. Transmission (R0) declines with increasing specific humidity from R0max to R0min. Three exemplar cities with low, intermediate, and high average specific humidity are shown across rows (Windhoek,

Antananarivo, and Lome, respectively). a-c, Proportion of the population infected (I/N) over time for the specified R0min and R0max values. d, Variation in peak size and timing when 1.0 < R0min < 1.5.

Nature Medicine | www.nature.com/naturemedicine